Robot Operating System

Sometime before 2007, the first pieces of what eventually would become ROS were beginning to come together at Stanford University.[16][17]  Eric Berger and Keenan Wyrobek, PhD students working in Kenneth Sailsbury's[18] robotics laboratory at Stanford, were leading the Personal Robotics Program.[19]  While working on robots to do manipulation tasks in human environments, the two students noticed that many of their colleagues were held back by the diverse nature of robotics: an excellent software developer might not have the hardware knowledge required, someone developing state of the art path planning might not know how to do the computer vision required. In an attempt to remedy this situation, the two students set out to make a baseline system that would provide a starting place for others in academia to build upon. In the words of Eric Berger, “something that didn’t suck, in all of those different dimensions”.[16]

In their first steps towards this unifying system, the two build the PR1 as a hardware prototype and began to work on software from it, borrowing the best practices from other early open source robotic software frameworks, particularly switchyard, a system that Morgan Quigley, another Stanford PhD student, had been working on in support of the STAIR (STanford Artificial Intelligence Robot)[20][21][22][23] by the Stanford Artificial Intelligence Laboratory.  Early funding of US$50,000 was provided by Joanna Hoffman and Alain Rossmann, which supported the development of the PR1. While seeking funding for further development,[24] Eric Berger and Keenan Wyrobek met Scott Hassan, the founder of Willow Garage, a technology incubator which was working on an autonomous SUV and a solar autonomous boat.  Hassan shared Berger and Wyrobek's vision of a “Linux for robotics”, and invited them to come and work at Willow Garage.  Willow Garage was started in January 2007, and the first commit of ROS code was made to SourceForge on the seventh of November, 2007.[25]

Willow Garage began developing the PR2 robot as a follow-up to the PR1, and ROS as the software to run it. Groups from more than twenty institutions made contributions to ROS, both the core software and the growing number of packages which worked with ROS to form a greater software ecosystem.[26][27] The fact that people outside of Willow were contributing to ROS (particularly from Stanford's STAIR project) meant that ROS was a multi-robot platform from the beginning. While Willow Garage had originally had other projects in progress, they were scrapped in favor of the Personal Robotics Program: focused on producing the PR2 as a research platform for academia and ROS as the open source robotics stack that would underlie both academic research and tech startups, much like the LAMP stack did for web-based startups.

In December 2008, Willow Garage met the first of their three internal milestones: continuous navigation for the PR2 over a period of two days and a distance of pi kilometers.[28] Soon after, an early version of ROS (0.4 Mango Tango)[29] was released, followed by the first RVIZ documentation and the first paper on ROS.[27]  In early summer, the second internal milestone: having the PR2 navigate the office, open doors, and plug itself it in, was reached.[30] This was followed in August by the initiation of the ROS.org website.[31] Early tutorials on ROS were posted in December,[32] preparing for the release of ROS 1.0, in January 2010.[33] This was Milestone 3: producing tons of documentation and tutorials for the enormous capabilities that Willow Garage's engineers had developed over the preceding 3 years.

Following this, Willow Garage achieved one of its longest held goals: giving away 10 PR2 robots to worthy academic institutions. This had long been a goal of the founders, as they felt that the PR2 could kick-start robotics research around the world. They ended up awarding eleven PR2s to different institutions, including University of Freiburg (Germany), Bosch, Georgia Tech, KU Leuven (Belgium), MIT, Stanford, TU Munich (Germany), UC Berkeley, U Penn, USC, and University of Tokyo (Japan).[34]  This, combined with Willow Garage's highly successful internship program[35] (run from 2008 to 2010 by Melonee Wise), helped to spread the word about ROS throughout the robotics world. The first official ROS distribution release: ROS Box Turtle, was released on March 2 of 2010, marking the first time that ROS was official distributed with a set of versioned packages for public use. These developments lead to the first drone running ROS,[36] the first autonomous car running ROS,[37] and the adaption of ROS for Lego Mindstorms.[38] With the PR2 Beta program well underway, the PR2 robot was officially released for commercial purchase on September 9, 2010.[39]

An image of Robot Operating System (ROS) running in Antarctica.

2011 was a banner year for ROS with the launch of ROS Answers, a Q/A forum for ROS users, on February 15;[40] the introduction of the highly successful Turtlebot robot kit on April 18;[41] and the total number of ROS repositories passing 100 on May 5.[42]  Willow Garage began 2012 by creating the Open Source Robotics Foundation (OSRF)[43] in April. The OSRF was immediately awarded a software contract by DARPA.[44] Later that year, the first ROSCon was held in St. Paul, MN,[45] the first book on ROS, ROS By Example,[46] was published, and the Baxter, first commercial robot to run ROS, was announced by Rethink Robotics.[47]  Soon after passing its fifth anniversary in November, ROS began running on every continent on December 3, 2012.[48]

In February 2013, the OSRF became the primary software maintainers for ROS,[49] foreshadowing the announcement in August that Willow Garage would be absorbed by its founders, Suitable Technologies.[50] At this point, ROS had released seven major versions (up to ROS Groovy[51]), and had users all over the globe.  This chapter of ROS development would be finalized when Clearpath Robotics took over support responsibilities for the PR2 in early 2014.[52]

In the years since OSRF took over primary development of ROS, a new version has been released every year,[51] while interest in ROS continues to grow. ROSCons have occurred every year since 2012, co-located with either ICRA or IROS, two flagship robotics conferences. Meetups of ROS developers have been organized in a variety of countries,[53][54][55] a number of ROS  books have been published,[56] and many educational programs initiated.[57][58] On September 1, 2014, NASA announced the first robot to run ROS in space: Robotnaut 2, on the International Space Station.[59] In 2017, the OSRF changed its name to Open Robotics.  Tech giants Amazon and Microsoft began to take an interest in ROS during this time, with Microsoft porting core ROS to Windows in September 2018,[60] followed by Amazon Web Services releasing RoboMaker in November.[61]

Perhaps the most important development of the OSRF/Open Robotics years thus far (not to discount the explosion of robot platforms which began to support ROS or the enormous improvements in each ROS version) was the proposal of ROS2, a significant API change to ROS which is intended to support real time programming, a wider variety of computing environments, and utilize more modern technology.[62] ROS2 was announced at ROSCon 2014,[63] the first commits to the ros2 repository were made in February 2015, followed by alpha releases in August 2015.[64]  The first distribution release of ROS2, Ardent Apalone, was released on December 8, 2017,[64] ushering in a new era of next-generation ROS development.

An image depicting the ROS equation: Plumbing + Tools + Capabilities + Ecosystem = ROS!

ROS was designed with open-source in mind, intending that users would be able to choose the configuration of tools and libraries which interacted with the core of ROS so that users could shift their software stacks to fit their robot and application area. As such, there is very little which is actually core to ROS, beyond the general structure within which programs must exist and communicate. In one sense, ROS is the underlying plumbing behind nodes and message passing. However, in reality, ROS is that plumbing, a rich and mature set of tools, a wide-ranging set of robot-agnostic capabilities provided by packages, and a greater ecosystem of additions to ROS.

ROS processes are represented as nodes in a graph structure, connected by edges called topics.[65] ROS nodes can pass messages to one another through topics, make service calls to other nodes, provide a service for other nodes, or set or retrieve shared data from a communal database called the parameter server. A process called the ROS Master[65] makes all of this possible by registering nodes to itself, setting up node-to-node communication for topics, and controlling parameter server updates. Messages and service calls do not pass through the master, rather the master sets up peer-to-peer communication between all node processes after they register themselves with the master. This decentralized architecture lends itself well to robots, which often consist of a subset of networked computer hardware, and may communicate with off-board computers for heavy computation or commands.

A node represents a single process running the ROS graph. Every node has a name, which it registers with the ROS master before it can take any other actions. Multiple nodes with different names can exist under different namespaces, or a node can be defined as anonymous, in which case it will randomly generate an additional identifier to add to its given name. Nodes are at the center of ROS programming, as most ROS client code is in the form of a ROS node which takes actions based on information received from other nodes, sends information to other nodes, or sends and receives requests for actions to and from other nodes.

Topics are named buses over which nodes send and receive messages.[66] Topic names must be unique within their namespace as well. To send messages to a topic, a node must publish to said topic, while to receive messages it must subscribe. The publish/subscribe model is anonymous: no node knows which nodes are sending or receiving on a topic, only that it is sending/receiving on that topic. The types of messages passed on a topic vary widely and can be user-defined. The content of these messages can be sensor data, motor control commands, state information, actuator commands, or anything else.

A node may also advertise services.[67] A service represents an action that a node can take which will have a single result. As such, services are often used for actions which have a defined beginning and end, such as capturing a single-frame image, rather than processing velocity commands to a wheel motor or odometer data from a wheel encoder. Nodes advertise services and call services from one another.

The parameter server[67] is a database shared between nodes which allows for communal access to static or semi-static information. Data which does not change frequently and as such will be infrequently accessed, such as the distance between two fixed points in the environment, or the weight of the robot, are good candidates for storage in the parameter server.

rviz[68] is a three-dimensional visualizer used to visualize robots, the environments they work in, and sensor data. It is a highly configurable tool, with many different types of visualizations and plugins.

rosbag[69] is a command line tool used to record and playback ROS message data. rosbag uses a file format called bags,[70] which log ROS messages by listening to topics and recording messages as they come in. Playing messages back from a bag is largely the same as having the original nodes which produced the data in the ROS computation graph, making bags a useful tool for recording data to be used in later development. While rosbag is a command line only tool, rqt_bag[71] provides a GUI interface to rosbag.

catkin[72] is the ROS build system, having replaced rosbuild[73] as of ROS Groovy. catkin is based on CMake, and is similarly cross-platform, open source, and language-independent.

The rosbash[74] package provides a suite of tools which augment the functionality of the bash shell. These tools include rosls, roscd, and roscp, which replicate the functionalities of ls, cd, and cp respectively. The ROS versions of these tools allow users to use ros package names in place of the filepath where the package is located. The package also adds tab-completion to most ROS utilities, and includes rosed, which edits a given file with the chosen default text editor, as well rosrun, which runs executes in ROS packages. rosbash supports the same functionalities for zsh and tcsh, to a lesser extent.

roslaunch[75] is a tool used to launch multiple ROS nodes both locally and remotely, as well as setting parameters on the ROS parameter server. roslaunch configuration files, which are written using XML can easily automate a complex startup and configuration process into a single command. roslaunch scripts can include other roslaunch scripts, launch nodes on specific machines, and even restart processes which die during execution.

• actionlib[76] provides a standardized interface for interfacing with preemtable tasks.
• nodelet[77] provides a way to run multiple algorithms in a single process.
• rosbridge[78] provides a JSON API to ROS functionalities for non-ROS programs.

• slam toolbox[79] provides full 2D SLAM and localization system.
• gmapping[80] provides a wrapper for OpenSlam's Gmapping algorithm for simultaneous localization and mapping.
• cartographer[81] provides real time 2D and 3D SLAM algorithms developed at Google.
• amcl[82] provides an implementation of adaptive Monte-Carlo localization.

• navigation[83] provides the capability of navigating a mobile robot in a planar environment.

• vision_opencv[84] is a meta-package which provides packages for integrating ROS with OpenCV.

• tf[85] provided a system for representing, tracking and transforming coordinate frames until ROS Hydro, when it was deprecated in favor of tf2.
• tf2[86] is the second generation of the tf library, and provides the same capabilities for ROS versions after Hydro.

• gazebo_ros_pkgs[87] is a meta-package which provides packages for integrating ROS with the Gazebo simulator.
• stage[88] provides an interface for the 2D Stage simulator.

ROS-Industrial[91] is an open-source project (BSD (legacy) / Apache 2.0 (preferred) license) that extends the advanced capabilities of ROS to manufacturing automation and robotics. The ROS-Industrial repository includes interfaces for common industrial manipulators, grippers, sensors, and device networks. It also provides software libraries for automatic 2D/3D sensor calibration, process path/motion planning, applications like Scan-N-Plan, developer tools like the Qt Creator ROS Plugin, and training curriculum that is specific to the needs of manufacturers. ROS-I is supported by an international Consortium of industry and research members. The project began as a collaborative endeavor between Yaskawa Motoman Robotics, Southwest Research Institute, and Willow Garage to support the use of ROS for manufacturing automation, with the GitHub repository being founded in January 2012 by Shaun Edwards (SwRI). Currently, the Consortium is divided into three groups; the ROS-Industrial Consortium Americas (led by SwRI and located in San Antonio, Texas), the ROS-Industrial Consortium Europe (led by Fraunhofer IPA and located in Stuttgart, Germany) and the ROS-Industrial Consortium Asia Pacific (led by Advanced Remanufacturing and Technology Centre (ARTC) and Nanyang Technological University (NTU) and located in Singapore).

The Consortia supports the global ROS-Industrial community by conducting ROS-I training, providing technical support and setting the future roadmap for ROS-I, as well as conducting pre-competitive joint industry projects to develop new ROS-I capabilities.[92]

• ABB, Adept, Fanuc, Motoman, and Universal Robots are supported by ROS-Industrial[93]
• Baxter[94] at Rethink Robotics, Inc.
• CK-9: Robotics Development Kit by Centauri Robotics, supports ROS
• HERB[95] developed at Carnegie Mellon University in Intel's personal robotics program
• Husky A200 robot developed (and integrated into ROS) by Clearpath Robotics[96]
• PR1 personal robot developed in Ken Salisbury's lab at Stanford[97]
• PR2 personal robot being developed at Willow Garage[98]
• Raven II Surgical Robotic Research Platform[99][100]
• Shadow Robot Hand[101] – A fully dexterous humanoid hand.
• STAIR I and II[102] robots developed in Andrew Ng's lab at Stanford
• SummitXL:[103] Mobile robot developed by Robotnik, an engineering company specialized in mobile robots, robotic arms, and industrial solutions with ROS architecture.
• Nao[104] humanoid: University of Freiburg's Humanoid Robots Lab[105] developed a ROS integration for the Nao humanoid based on an initial port by Brown University[106][107]
• UBR1[108][109] developed by Unbounded Robotics, a spin-off of Willow Garage.
• ROSbot: autonomous robot platform by Husarion[110]
• Webots: robot simulator integrating a complete ROS programming interface.[111]

• BeagleBoard. The robotics lab of the Katholieke Universiteit Leuven, Belgium[112] has ported ROS to the Beagleboard
• Sitara ARM Processors have support for the ROS package as part of the official Linux SDK[113].
• Raspberry Pi: image of ubuntu Mate with ROS[114] by Ubiquity Robotics; installation guide for Raspbian[115]